Multi-Robot Path Planning Based on Learning Classifier System with Policy Gradient Reinforcement Learning and Support Vector Machine (PGRL-SVM)

Abstract

This paper presented a novel approach to solving the problem of robot avoidance collision planning. A Learning Classifier System is a accuracy-based machine learning system using gradient descent that combines covering operator and genetic algorithm. The covering operator is responsible for adjusting precision and large search space according to some reward obtained from the environment. The genetic algorithm acts as an innovation discovery component which is responsible for discovering new better path planning rules. The advantages of this approach are its accuracy-based representation, that can be easily reduce learning space,online learning ability,robustness due to the use of genetic algorithm.